For quality engineers at contract machine shops

The part fits.
The wrench doesn't. Find the blocked wrench. Before the print.

A perfect fit can still block the job. FixtureFit uses Gemini 4 Argon to design and repair the holding fixture, while an independent geometric checker measures whether the modelled tool clears it, before you order a test print.

Try the live demo Start a job

Live: one click to the sample job, one click on Repair (two Argon forks, about 70–90 s). 3 live repairs per visitor per hour; after that, the labelled reference run. New to fixtures? Explained simply →

The model writes the design. FixtureFit's checker gives the clearance verdict. The engineer signs off.

Measured example · clearance study

CAD-style illustration · recorded results
Socket on an extension · enters from the leftClears the posts with margin
After one Argon repair · socket outside Ø28 mm
+1.97 mmPASS
Required margin1.5 mm
Try the live demo →

Front-end illustration — not a live clearance check. The slider redraws the socket; it does not run the checker. The workbench re-runs the real checker on any socket.

A CAD-style illustration of the sample job: a metal housing held in a printed holder with two posts on its left side. A socket on an extension comes in from the left between the posts to the bolt. Recorded results from FixtureFit's checker: Draft 1 · standard socket outside Ø18 mm, +3.98 mm, PASS. Same Draft 1 · bigger socket outside Ø28 mm, −1.03 mm, FAIL: the socket hits the holder by 1.03 mm. After one Argon repair · socket outside Ø28 mm, +1.97 mm, PASS against the 1.5 mm margin. The part never moves.

One example, three recorded states.

The checker's numbers from the run records · separate from the illustration above

  1. Draft 1 · standard socket outside Ø18 mm

    +3.98 mm · PASS

  2. Same Draft 1 · bigger socket outside Ø28 mm

    −1.03 mm · FAIL

  3. After one Argon repair · socket outside Ø28 mm

    +1.97 mm · PASS

Required margin 1.5 mm. Negative = the wrench hits the holder by that much. The repair was measured live 3 times: 3 of 3 passed in 70–87 s, from the request to the checker's verdict. One fixture family; not a promise for every design.

The interactive model · explore the geometry

Same part. Different access.

Look at the clash from your own angle. Switch between the three recorded states, look down the tool path, compare the two holders, or redraw the socket bigger or smaller. The part stays exactly where it is.

Figure 02 · tool access studyIso
Same Draft 1 · bigger socket outside Ø28 mm−1.03 mm · FAILChecker's recorded result · not recalculated here

Draft 1 with the Ø28 socket. Front-end illustration, not a live clearance check.Get a real number from the checker in the live demo →

The part fits. That's not the whole job.

The drawing holds the part.
The operator needs a way in.

A 3D-printed holder can grip the part perfectly and still block the operator's wrench. The drawing doesn't show the socket coming in from the side.

So the clash is found at the machine, after the print: the holder is ready, the part is clamped, and the wrench can't reach the bolt. Another design, another print, another machine slot.

Panel 1, Hold the part: an operator tightens a bolt on a metal part held in a blue 3D-printed fixture.Panel 2, Oops, blocked: the wrench hits the fixture; minus 1.03 mm means the wrench hits the fixture by that much.Panel 3, AI redesigns: watch the worker's clip, design a fixture, test and fix, you approve.Panel 4, Safe to try: the redesigned fixture leaves room; plus 1.97 mm means the wrench clears by that much.
Illustration The workflow in four panels. The two numbers are the checker's recorded results for the sample job: −1.03 mm with the socket outside Ø28 mm, +1.97 mm after the repair. Open full size

From operator motion to checked geometry

Design around the work.
Not just the part.

Gemini 4 Argon watches the operator, writes the holder and repairs it. FixtureFit's own checker sweeps the tool and measures the gap. The blocked wrench shows up on screen, not after the print.

AI-generated illustration (Gemini Omni)
AI-generated illustration (Gemini Omni). Not footage of a real shop: it shows the idea; the numbers on this page come from FixtureFit's checker.

  1. 1 · Watch and confirm

    Gemini 4 Argon watches a short clip of the operator and writes keep-clear rules with timestamps: which tool, from which side, onto which face. No millimetre is read from pixels. The engineer confirms each rule.

    operator motion → engineer-confirmed rules

  2. 2 · Draft the holder

    Argon writes the whole first fixture as CAD code in one pass, from the confirmed rule, the part and the tool. In the sample job it took 112 s. A draft is a starting point, not a verdict.

  3. 3 · Check the tool path

    FixtureFit's checker runs outside Argon's sandbox, on its own copy of the part, tool and spec. It sweeps the socket along its path and measures the smallest gap in millimetres.

    the checker measures · the model does not decide

  4. 4 · Repair

    When a change makes the wrench hit, Argon gets the checker's numbers inside Gemini Managed Agents: one saved agent, two forks at once (move a wall / cut a relief), sandbox network off. A failing fork gets the checker's verdict back and tries again; the checker keeps the fork that clears.

  5. 5 · Engineer review

    Every repair stops on a sign-off request. Only a fixture that passed can be approved; then a test-prototype order (Stripe test mode).

The measured example

Three states.
Three numbers. One checker.

Each number belongs to one state of one sample job: a sensor housing with an M8 bolt in its left face, torqued with a socket on an extension. A number is never carried from one state to the next.

A pre-registered baseline on other jobs hit its ceiling (every approach passed every job), so I make no claim that the repair loop makes Argon better. The claim is the independent check. See the evals.

State 1 · socket outside Ø18 mm

Draft 1

+3.98 mm PASS

Argon's first holder, written in one pass. The standard socket clears both posts.

runs/draft1/report.json
State 2 · socket outside Ø28 mm

Draft 1, bigger socket

−1.03 mm FAIL

A bigger socket arrives. Same holder: the wrench hits it by 1.03 mm.

runs/forks/summary.json (seed)
State 3 · socket outside Ø28 mm

After one Argon repair

+1.97 mm PASS

Two Argon forks in Managed Agents; the checker kept the relief cut. 3 of 3 live runs passed in 70–87 s.

runs/stage-trials.json

Clear responsibilities. An independent verdict.

The model proposes.
The checker decides.

Google's models read, write and repair. FixtureFit measures clearance. The engineer decides whether to go ahead.

Read · write · repair

Gemini 4 Argon

Reads the operator clip, writes Draft 1 as CAD code, and repairs the holder inside Managed Agents.

Controlled sessions

Gemini Managed Agents

One saved agent, two forks at the same time. Sandbox network off. Each repair pauses for the engineer.

Helper line

Gemini 3.8 Flash

One plain sentence about a failing check.

not part of the verdict
Sample input

Gemini Omni

Made the sample operator clip and the film above.

AI-generated, labelled
Outside the AI sandbox

FixtureFit's independent checker

Re-imports every fixture, sweeps the tool along its path and measures clearance in millimetres; also part overlap, locating points, the camera's keep-out zone, one closed solid and the printer's envelope. The checker, not an AI model, always gives the verdict. A model's description of its own design is not a passing check.

Google models: generateFixtureFit: checkEngineer: sign off

What leaves FixtureFit

Geometry to test.
A check to review.

The result has passed a clearance check on the modelled tool path. It is a geometrically screened prototype candidate; the engineer signs off.

fixture.pySTEP · STL
The files the workbench exports

Screened fixture geometry

The fixture program (CAD code) and its STEP and STL, ready for the engineer's review before a test prototype is ordered.

Clearance check · modelled pathØ28
+1.97 mmPASS

Required 1.5 mm · collision ✓ · datums ✓ · keep-out ✓ · watertight ✓ · envelope ✓

Not evaluated: clamping force, stability under load, material

Report summary · the sample job after the repair

The clearance report

The checker's result next to the geometry, with what was not evaluated, the checker's version and a hash of every input. Not an AI-written promise that the tool will fit.

A modelled clearance result is not physical validation, a strength assessment, a manufacturability review, or proof of unrestricted tool access.

At the contract machine shop

For the engineer who has to make it work.

For quality and manufacturing engineers designing one-off 3D-printed fixtures: the holders that keep a part still while an operator bolts, probes or inspects it.

Keep your judgement in the loop. Put the operator's tool path in the design loop.

  • Bolting
  • Probing
  • Inspection

Straightforward pricing

Screen the geometry.
Then test the print.

Pay for a candidate that passed the check and your engineer approved. Drafts and fixes that fail are not charged. Pricing details

Per screened fixture candidate

$15per screened fixture candidate, plus the test prototype at cost.

Includes the fixture program, STEP and STL, the clearance report and the sign-off record.

Payments run in Stripe test mode today: no real money moves and nothing is printed.

Find the blocked wrench.
Before the print.

The part needs to stay put. The wrench needs a way in. Check both before you order the test prototype.

Try the live demo